1
00:00:00,000 --> 00:00:03,050


2
00:00:03,050 --> 00:00:04,770
He wasn't taping yet.

3
00:00:04,770 --> 00:00:07,600
OK, so thank you for coming.

4
00:00:07,600 --> 00:00:08,880
My pleasure.

5
00:00:08,880 --> 00:00:12,150
So why don't you tell me
what you're doing right now?

6
00:00:12,150 --> 00:00:15,660
What I'm doing right
now in the field of AI

7
00:00:15,660 --> 00:00:19,590
is concerned with
cases and examples

8
00:00:19,590 --> 00:00:22,950
and how one reasons with them.

9
00:00:22,950 --> 00:00:27,840
Particularly interested
in the problem of how

10
00:00:27,840 --> 00:00:31,590
one reasons with concepts
that are messy in the sense

11
00:00:31,590 --> 00:00:35,150
that they're not well-defined
say by logical definition

12
00:00:35,150 --> 00:00:37,200
to necessary
intuition conditions,

13
00:00:37,200 --> 00:00:40,080
concepts that have
exceptions and holes in them,

14
00:00:40,080 --> 00:00:43,230
and concepts that
change, and the role

15
00:00:43,230 --> 00:00:45,630
of examples in cases
in representing

16
00:00:45,630 --> 00:00:47,480
those sorts of things.

17
00:00:47,480 --> 00:00:48,750
OK.

18
00:00:48,750 --> 00:00:50,760
That's what drives me.

19
00:00:50,760 --> 00:00:53,310
I love examples and
cases and hypotheticals.

20
00:00:53,310 --> 00:00:56,610
Fabulous, and how did you get
into the field originally?

21
00:00:56,610 --> 00:01:00,930
Well, came up pretty classically
through the mathematics route.

22
00:01:00,930 --> 00:01:04,260
I was an undergraduate
mathematics major at Brown

23
00:01:04,260 --> 00:01:06,480
and I loved mathematics.

24
00:01:06,480 --> 00:01:09,420
I also like thinking
about the problem of how

25
00:01:09,420 --> 00:01:13,260
you get good at mathematics but
nonetheless, eventually ended

26
00:01:13,260 --> 00:01:18,420
up in the MIT math department
as a regular graduate student.

27
00:01:18,420 --> 00:01:21,270
And at some point
got sort of bitchy

28
00:01:21,270 --> 00:01:23,760
about continuing to
be a graduate student.

29
00:01:23,760 --> 00:01:29,850
And took a year off to go work
at Lincoln Labs supposedly

30
00:01:29,850 --> 00:01:35,730
to get some experience on
stability problems etc.

31
00:01:35,730 --> 00:01:39,870
but when I got to the labs I
ended up being put on a project

32
00:01:39,870 --> 00:01:41,790
that I found quite boring.

33
00:01:41,790 --> 00:01:43,890
And I resorted to
the sorts of things

34
00:01:43,890 --> 00:01:45,962
I did high school which
was go to the library

35
00:01:45,962 --> 00:01:47,920
and think about things
I wanted to think about.

36
00:01:47,920 --> 00:01:48,690
How terrible.

37
00:01:48,690 --> 00:01:49,900
How terrible.

38
00:01:49,900 --> 00:01:54,420
And I proposed a project
that would essentially

39
00:01:54,420 --> 00:01:57,990
build what we would call today
Mathematicians Workbench, that

40
00:01:57,990 --> 00:02:01,710
would be a repository for
information like theorems

41
00:02:01,710 --> 00:02:05,760
and definitions and
counterexamples in mathematics

42
00:02:05,760 --> 00:02:09,360
and all the interconnections
amongst them

43
00:02:09,360 --> 00:02:11,380
so that you could ask questions.

44
00:02:11,380 --> 00:02:14,970
Well, if I'm looking
at this theorem can

45
00:02:14,970 --> 00:02:17,070
you tell me whether I
can drop a precondition

46
00:02:17,070 --> 00:02:20,295
and if so, point me to the
theorem and the proof, if not

47
00:02:20,295 --> 00:02:21,420
show me the counterexample.

48
00:02:21,420 --> 00:02:22,200
Is this a program?

49
00:02:22,200 --> 00:02:23,580
Oh by the way--

50
00:02:23,580 --> 00:02:32,040
I wanted it-- it's essentially
a hypertext web-like environment

51
00:02:32,040 --> 00:02:36,150
that would let you
navigate very, very

52
00:02:36,150 --> 00:02:41,610
freely among this highly
interconnected knowledge

53
00:02:41,610 --> 00:02:42,780
base of mathematics.

54
00:02:42,780 --> 00:02:46,860
And so I wanted to first decide
what should be in the knowledge

55
00:02:46,860 --> 00:02:50,190
base and then build a
program that would let you

56
00:02:50,190 --> 00:02:54,210
you know, swing
around it very freely.

57
00:02:54,210 --> 00:02:57,810
And that's how it started
off and I proposed that

58
00:02:57,810 --> 00:02:59,400
to my supervisors at MIT--

59
00:02:59,400 --> 00:03:00,360
at Lincoln Labs.

60
00:03:00,360 --> 00:03:03,490
And they said, well no, we're
not funded to let you do that.

61
00:03:03,490 --> 00:03:07,590
But back on campus some
people in the math department,

62
00:03:07,590 --> 00:03:09,270
particularly they
chair, heard about it

63
00:03:09,270 --> 00:03:11,790
and said, why don't you go
talk to Marvin and Seymour

64
00:03:11,790 --> 00:03:13,350
about this idea.

65
00:03:13,350 --> 00:03:16,210
And Seymour in particular
who's a professor,

66
00:03:16,210 --> 00:03:18,840
Seymour Papert who's a
professor of mathematics

67
00:03:18,840 --> 00:03:22,345
said, well, why don't you come
back to do this as your thesis?

68
00:03:22,345 --> 00:03:23,790


69
00:03:23,790 --> 00:03:24,900
What year was this?

70
00:03:24,900 --> 00:03:35,670
This was-- I was at Lincoln Labs
in '74-'75 and so I was having

71
00:03:35,670 --> 00:03:38,160
these conversations with
them I guess the summer

72
00:03:38,160 --> 00:03:40,752
of '74, the fall of '74.

73
00:03:40,752 --> 00:03:42,210
I'd have to check
my dates exactly.

74
00:03:42,210 --> 00:03:44,760
Were they both the-- heading
the AI lab at the time?

75
00:03:44,760 --> 00:03:48,780
Yes, but I didn't know them
so much through the AI lab

76
00:03:48,780 --> 00:03:52,035
as through my home base
in the math department.

77
00:03:52,035 --> 00:03:53,220
OK.

78
00:03:53,220 --> 00:03:56,520
There's an awful lot
of AI that has happened

79
00:03:56,520 --> 00:03:58,620
in the math department at MIT.

80
00:03:58,620 --> 00:04:00,310
There's a really
strong tradition.

81
00:04:00,310 --> 00:04:02,670
And so-- and of
the mathematicians

82
00:04:02,670 --> 00:04:05,710
there are not scared to think
about things in new ways

83
00:04:05,710 --> 00:04:08,160
and so in fact the person
who probably pointed me off

84
00:04:08,160 --> 00:04:13,390
in this direction is a classical
functional analyst type.

85
00:04:13,390 --> 00:04:16,779
And so, so anyway, Seymour
said, why don't you

86
00:04:16,779 --> 00:04:20,500
come back to campus and
do this as your thesis?

87
00:04:20,500 --> 00:04:22,570
And whereas I thought
I was going to do

88
00:04:22,570 --> 00:04:23,720
a thesis on something else.

89
00:04:23,720 --> 00:04:26,620
And I thought gee, my
hobby is my thesis,

90
00:04:26,620 --> 00:04:28,210
what could be better?

91
00:04:28,210 --> 00:04:34,390
And so it's essentially a swing
over to the AI side which is--

92
00:04:34,390 --> 00:04:37,180
now you would call it sort of
an [INAUDIBLE] representation

93
00:04:37,180 --> 00:04:39,710
or an environmental
kind of thing.

94
00:04:39,710 --> 00:04:44,920
And one of the important
things I thought

95
00:04:44,920 --> 00:04:48,700
to capture in this
environment were the examples,

96
00:04:48,700 --> 00:04:52,270
how one counterexample or one
example is build from another.

97
00:04:52,270 --> 00:04:55,780
And I thought that was
as much an ingredient

98
00:04:55,780 --> 00:04:58,000
of mathematical knowledge
as the theorems,

99
00:04:58,000 --> 00:05:00,767
how one theorem is proved
from another theorem.

100
00:05:00,767 --> 00:05:02,350
And of course there's
interconnections

101
00:05:02,350 --> 00:05:03,900
between the theorems
and the examples

102
00:05:03,900 --> 00:05:06,280
so my initial work
was on building up

103
00:05:06,280 --> 00:05:08,800
this representation
that made room

104
00:05:08,800 --> 00:05:13,840
for these sort of non-logical
or non-Sunday school mathematics

105
00:05:13,840 --> 00:05:15,740
kinds of objects.

106
00:05:15,740 --> 00:05:20,056
And so--

107
00:05:20,056 --> 00:05:22,810
Had any other stuff been
done like that before?

108
00:05:22,810 --> 00:05:28,360
No, I mean, there certainly were
compendia of counterexamples

109
00:05:28,360 --> 00:05:29,560
in mathematics.

110
00:05:29,560 --> 00:05:32,400
There certainly were
books of theorems,

111
00:05:32,400 --> 00:05:35,440
I mean obviously classical
mathematics is all about that.

112
00:05:35,440 --> 00:05:37,850
And there were a
few instances of how

113
00:05:37,850 --> 00:05:42,220
to think about
mathematics such as Polya.

114
00:05:42,220 --> 00:05:48,850
But a representational scheme
in an environment like this, no.

115
00:05:48,850 --> 00:05:51,730
And so there was the
representational substrate

116
00:05:51,730 --> 00:05:54,430
and epistemology of
mathematics if you want.

117
00:05:54,430 --> 00:05:59,260
A way to use that representation
for the mathematician

118
00:05:59,260 --> 00:06:01,600
who knows what she wants.

119
00:06:01,600 --> 00:06:05,260
And then I had this idea
that with all the structure

120
00:06:05,260 --> 00:06:08,740
you should be able
to describe what

121
00:06:08,740 --> 00:06:12,070
it means to understand the
body of mathematics well,

122
00:06:12,070 --> 00:06:16,000
and then to help someone
else learn how to understand

123
00:06:16,000 --> 00:06:17,430
a body of mathematics.

124
00:06:17,430 --> 00:06:20,140
And this is more
Polya-esque in the sense

125
00:06:20,140 --> 00:06:23,350
that if someone-- if you learn
a theorem you should ask well

126
00:06:23,350 --> 00:06:26,080
is the converse true?

127
00:06:26,080 --> 00:06:29,870
And if it's not true well,
what's the counterexample that

128
00:06:29,870 --> 00:06:30,910
disproves it?

129
00:06:30,910 --> 00:06:33,490
You should learn to ask
those kinds of questions

130
00:06:33,490 --> 00:06:39,160
to weave together this landscape
of mathematical knowledge

131
00:06:39,160 --> 00:06:41,590
and to know your way around
it and know what's important

132
00:06:41,590 --> 00:06:44,150
and know what's just
merely a stepping stone

133
00:06:44,150 --> 00:06:47,200
and know how certain
examples can tie together

134
00:06:47,200 --> 00:06:48,490
different concepts.

135
00:06:48,490 --> 00:06:51,700
How some ideas permeate
things, the heuristic

136
00:06:51,700 --> 00:06:53,300
content of mathematics.

137
00:06:53,300 --> 00:06:56,586
So for instance to
try extreme points

138
00:06:56,586 --> 00:06:58,030
you know, that's in programming.

139
00:06:58,030 --> 00:07:01,480
You know, and there's a lot
of this knowledge that's

140
00:07:01,480 --> 00:07:02,360
very important.

141
00:07:02,360 --> 00:07:06,790
And so it like this should
all be part of the bundle

142
00:07:06,790 --> 00:07:10,630
and it has a landscape and
it has ways to navigate it.

143
00:07:10,630 --> 00:07:14,260
And so that became my thesis
and that my official thesis

144
00:07:14,260 --> 00:07:17,750
advisor was Seymour Papert
and all my committee

145
00:07:17,750 --> 00:07:22,850
my co-advisor was Marvin
Minsky and it was a lot of fun.

146
00:07:22,850 --> 00:07:23,350
Cool.

147
00:07:23,350 --> 00:07:26,420
What was the environment
like at MIT at the time?

148
00:07:26,420 --> 00:07:28,580
Who were you collaborating with?

149
00:07:28,580 --> 00:07:31,910
Well, I wasn't collaborating
with anybody on my thesis.

150
00:07:31,910 --> 00:07:34,850
I mean, that was pretty
much a solo enterprise.

151
00:07:34,850 --> 00:07:38,600
The MIT style seems
to be to set you loose

152
00:07:38,600 --> 00:07:41,040
and give you all the
freedom you want.

153
00:07:41,040 --> 00:07:42,680
And for me that
was great because I

154
00:07:42,680 --> 00:07:44,450
knew what I wanted to do.

155
00:07:44,450 --> 00:07:46,040
In retrospect, it
would have been

156
00:07:46,040 --> 00:07:50,240
better to have had more
critical interactions

157
00:07:50,240 --> 00:07:53,160
and perhaps more feedback on
the goodness of the ideas.

158
00:07:53,160 --> 00:07:55,340
But when I was in the,
in the throes of it

159
00:07:55,340 --> 00:07:57,950
I quite liked the freedom I had.

160
00:07:57,950 --> 00:08:01,673
So I guess my contacts were
some of the mathematicians,

161
00:08:01,673 --> 00:08:03,090
some people like
Ken Hoffman who's

162
00:08:03,090 --> 00:08:06,030
an analyst, and
people in the AI lab,

163
00:08:06,030 --> 00:08:08,960
particularly Marvin and Seymour.

164
00:08:08,960 --> 00:08:12,535
What was that atmosphere
like at the AI lab then?

165
00:08:12,535 --> 00:08:17,360
I don't know, pretty
freewheeling, pretty tolerant

166
00:08:17,360 --> 00:08:23,090
of all kinds of new ideas,
very individualistic.

167
00:08:23,090 --> 00:08:26,285
I think a lot of strong,
smart individuals

168
00:08:26,285 --> 00:08:29,170
you know who could sort of--

169
00:08:29,170 --> 00:08:32,960
it wasn't a project
oriented place say

170
00:08:32,960 --> 00:08:34,520
as much as some
place like Stanford

171
00:08:34,520 --> 00:08:37,340
was which at least to me
appeared to be more oriented

172
00:08:37,340 --> 00:08:40,610
around projects that involved
a lot of graduate students

173
00:08:40,610 --> 00:08:43,400
working, there seem to
be more solo enterprises.

174
00:08:43,400 --> 00:08:46,820
And of course, I spent
my time in a way in three

175
00:08:46,820 --> 00:08:48,380
different entities at MIT.

176
00:08:48,380 --> 00:08:52,190
There was the AI lab, there
was the math department,

177
00:08:52,190 --> 00:08:54,370
and there was another entity
which no longer exists

178
00:08:54,370 --> 00:08:56,120
called the Division
For Study and Research

179
00:08:56,120 --> 00:09:01,700
in Education which was trying
to marry lots of disciplines

180
00:09:01,700 --> 00:09:03,635
such as philosophy,
linguistics, and AI

181
00:09:03,635 --> 00:09:10,610
and mathematics that had along
the perspective of education

182
00:09:10,610 --> 00:09:11,575
and--

183
00:09:11,575 --> 00:09:13,700
but it was broader than
just teaching other people.

184
00:09:13,700 --> 00:09:15,830
It was more of an
AI cognitive science

185
00:09:15,830 --> 00:09:18,620
kind of orientation
and understanding

186
00:09:18,620 --> 00:09:20,910
how thought goes in
these other disciplines.

187
00:09:20,910 --> 00:09:23,180
So there were people
from [INAUDIBLE]

188
00:09:23,180 --> 00:09:25,545
from the architecture
school and there

189
00:09:25,545 --> 00:09:27,845
was Jeanne Bamberger
from music and it

190
00:09:27,845 --> 00:09:29,540
was-- that was a very
interesting place.

191
00:09:29,540 --> 00:09:30,373
It was a cool place.

192
00:09:30,373 --> 00:09:33,083
It was too bad that the
enterprise didn't continue.

193
00:09:33,083 --> 00:09:35,000
So speaking of those
kinds of things what kind

194
00:09:35,000 --> 00:09:38,060
of collaborations do you think
exist between different fields

195
00:09:38,060 --> 00:09:39,600
and Artificial Intelligence?

196
00:09:39,600 --> 00:09:43,260
Or where potential is
for that to happen?

197
00:09:43,260 --> 00:09:48,450
Well, I think some non-computer
science disciplines

198
00:09:48,450 --> 00:09:52,230
have more intimate
collaborations

199
00:09:52,230 --> 00:09:54,460
with AI than others.

200
00:09:54,460 --> 00:09:59,640
So for instance I think things
like biology, neuroscience,

201
00:09:59,640 --> 00:10:05,070
chemistry, VLSI,
CAD/CAM engineering,

202
00:10:05,070 --> 00:10:08,790
things like that have had
quite an intimate relationship

203
00:10:08,790 --> 00:10:12,750
with AI and that use
the AI techniques.

204
00:10:12,750 --> 00:10:16,350
Other disciplines are
more at a distance

205
00:10:16,350 --> 00:10:22,120
such as law and maybe ethics and
some of the softer disciplines.

206
00:10:22,120 --> 00:10:24,610
But I think the interaction
there is just as interesting.

207
00:10:24,610 --> 00:10:28,410
So in fact, after I had done
this work in mathematics

208
00:10:28,410 --> 00:10:30,000
I found out that
most of my colleagues

209
00:10:30,000 --> 00:10:34,080
were math-a-phobic or
really didn't want to get

210
00:10:34,080 --> 00:10:35,430
into it to the degree I did.

211
00:10:35,430 --> 00:10:38,460
And I felt that the
phenomenon of examples

212
00:10:38,460 --> 00:10:39,960
and how you reason
with them and how

213
00:10:39,960 --> 00:10:43,870
you build them should be
present in other disciplines.

214
00:10:43,870 --> 00:10:45,660
So for instance, the
way I wrote programs.

215
00:10:45,660 --> 00:10:48,900
Very often if I had to
write a new search program I

216
00:10:48,900 --> 00:10:51,000
would look up the last
one I wrote and tweak it

217
00:10:51,000 --> 00:10:53,670
a bit, which is essentially a
case based for example based

218
00:10:53,670 --> 00:10:56,190
method of retrieving
a good enough example

219
00:10:56,190 --> 00:10:57,600
and then pushing it further.

220
00:10:57,600 --> 00:11:00,060
That's also the way you
can go counterexample.

221
00:11:00,060 --> 00:11:03,990
So I tried to look at other
disciplines besides mathematics

222
00:11:03,990 --> 00:11:07,500
where the sort of problem
solving behavior was useful.

223
00:11:07,500 --> 00:11:08,670
And the one I--

224
00:11:08,670 --> 00:11:10,500
it's useful in
linguistics, but the one

225
00:11:10,500 --> 00:11:15,060
I really love and have since
spent most of my career in

226
00:11:15,060 --> 00:11:19,260
is the legal domain because
at least Anglo-American law

227
00:11:19,260 --> 00:11:23,490
and other types of law as
well but to a lesser extent

228
00:11:23,490 --> 00:11:25,120
are based on cases.

229
00:11:25,120 --> 00:11:27,090
In other words examples.

230
00:11:27,090 --> 00:11:31,020
And what's interesting
about cases and examples

231
00:11:31,020 --> 00:11:34,950
in legal domains is there's a
difference between real cases

232
00:11:34,950 --> 00:11:37,200
and there's no hard definition
of that and essentially

233
00:11:37,200 --> 00:11:40,200
something that has been
brought to the legal system

234
00:11:40,200 --> 00:11:41,400
and hypotheticals.

235
00:11:41,400 --> 00:11:44,070
Hypotheticals are made up
cases, very good [INAUDIBLE]

236
00:11:44,070 --> 00:11:46,950
experiments streamline
kind of examples.

237
00:11:46,950 --> 00:11:48,960
And they're more
like counterexamples

238
00:11:48,960 --> 00:11:50,700
in the sense of mathematics.

239
00:11:50,700 --> 00:11:53,748
And they get used in
Socratic dialogue,

240
00:11:53,748 --> 00:11:55,290
the back and forth
when you're trying

241
00:11:55,290 --> 00:11:58,140
to test your ideas or the
professors trying to push you

242
00:11:58,140 --> 00:12:00,570
around to see whether
what you're talking about,

243
00:12:00,570 --> 00:12:04,080
or an oral argument in
appellate situations

244
00:12:04,080 --> 00:12:06,960
where someone's trying to see
will your arguments stand up

245
00:12:06,960 --> 00:12:09,700
if the situation is
slightly different.

246
00:12:09,700 --> 00:12:13,770
So I think there's a nice
potential relationship

247
00:12:13,770 --> 00:12:17,075
with disciplines like
the law and a few of us

248
00:12:17,075 --> 00:12:17,980
have pursued it.

249
00:12:17,980 --> 00:12:20,263
But it's not as
widespread as I'd

250
00:12:20,263 --> 00:12:21,930
like because I think
it's a rich domain.

251
00:12:21,930 --> 00:12:26,370
So what can AI do
for the field of law?

252
00:12:26,370 --> 00:12:27,660
Well I think AI can do for--

253
00:12:27,660 --> 00:12:29,470
What kind of applications
are there in the near future?

254
00:12:29,470 --> 00:12:30,845
Well, that's a
different question

255
00:12:30,845 --> 00:12:32,490
or that's a sub-question.

256
00:12:32,490 --> 00:12:36,030
I think in the
general sense AI can

257
00:12:36,030 --> 00:12:41,460
do for law what it can do for
many areas in that it gives

258
00:12:41,460 --> 00:12:43,512
these other people
in these other areas

259
00:12:43,512 --> 00:12:44,970
who want to think
about their field

260
00:12:44,970 --> 00:12:47,410
a framework for thinking
about their field.

261
00:12:47,410 --> 00:12:49,920
So if you're interested in
describing problem solving

262
00:12:49,920 --> 00:12:52,240
or what it means to
think like a lawyer

263
00:12:52,240 --> 00:12:54,645
and to paraphrase Kingsfeld
from the pictures.

264
00:12:54,645 --> 00:12:57,180


265
00:12:57,180 --> 00:13:00,300
The people in legal
philosophy or who

266
00:13:00,300 --> 00:13:04,380
look at legal reasoning on
the legal side I don't think

267
00:13:04,380 --> 00:13:06,960
have a very rich vocabulary
for describing what it

268
00:13:06,960 --> 00:13:09,300
means to think like a lawyer.

269
00:13:09,300 --> 00:13:12,060
And for instance the
doctrine of precedent

270
00:13:12,060 --> 00:13:15,810
says that similar cases
should be decided similarly

271
00:13:15,810 --> 00:13:17,520
but there's no
further explanation

272
00:13:17,520 --> 00:13:20,340
of what it means for
two cases to be similar.

273
00:13:20,340 --> 00:13:24,990
Well in fact, there's lots of
ways to describe similarities.

274
00:13:24,990 --> 00:13:31,260
And AI gives you a handle on
making quite precise what sort

275
00:13:31,260 --> 00:13:34,110
of similarity you're interested
in and how to manipulate it

276
00:13:34,110 --> 00:13:36,070
and how to change it.

277
00:13:36,070 --> 00:13:39,750
And so AI studies
of legal reasoning

278
00:13:39,750 --> 00:13:43,890
can shed light on the
jurisprudence of a case based

279
00:13:43,890 --> 00:13:47,370
reasoning and what it means
for cases to be similar

280
00:13:47,370 --> 00:13:51,870
and how do you create those
nasty hypotheticals anyway?

281
00:13:51,870 --> 00:13:57,690
And that not only is kind of a
philosophical jurisprudential

282
00:13:57,690 --> 00:14:00,870
enterprise, but it also could
be an educational enterprise

283
00:14:00,870 --> 00:14:05,070
if you want to teach law
students how to push back

284
00:14:05,070 --> 00:14:08,580
with good hypotheticals or
you want people in practice

285
00:14:08,580 --> 00:14:12,630
to think about how can you
imagine situations that

286
00:14:12,630 --> 00:14:14,708
might get you into trouble?

287
00:14:14,708 --> 00:14:16,500
You need to think about
hypotheticals, well

288
00:14:16,500 --> 00:14:19,420
how should you come up
with those kinds of cases?

289
00:14:19,420 --> 00:14:23,430
So I think the kind of work
that we do in AI demystifies

290
00:14:23,430 --> 00:14:27,550
some of those processes
which seem quite magical.

291
00:14:27,550 --> 00:14:29,940
I mean, if you think back
to your undergraduate years

292
00:14:29,940 --> 00:14:33,390
taking real analysis how
did those people come up

293
00:14:33,390 --> 00:14:35,340
with those wild counterexamples?

294
00:14:35,340 --> 00:14:36,840
It seemed like magic.

295
00:14:36,840 --> 00:14:40,440
Well it's not magic, at
least it's not all magic.

296
00:14:40,440 --> 00:14:42,320
And the there are
some models for how

297
00:14:42,320 --> 00:14:44,360
to do that that can
shed light on it

298
00:14:44,360 --> 00:14:47,300
and those are AI's contribution,
not only to the field

299
00:14:47,300 --> 00:14:49,320
but to how to teach the field.

300
00:14:49,320 --> 00:14:52,640
Well, so in the future do
you see machines or computers

301
00:14:52,640 --> 00:14:57,180
as being able to give
advice to lawyers

302
00:14:57,180 --> 00:14:58,860
or politicians
like in a courtroom

303
00:14:58,860 --> 00:14:59,985
or in a government setting?

304
00:14:59,985 --> 00:15:03,150
I see them as able assistants.

305
00:15:03,150 --> 00:15:05,610
I don't ever see them
in the role of Judge.

306
00:15:05,610 --> 00:15:08,010
I think that's silly
because the law is

307
00:15:08,010 --> 00:15:10,240
much more complex than subtle.

308
00:15:10,240 --> 00:15:10,740
Right.

309
00:15:10,740 --> 00:15:12,730
It's a very subtle enterprise.

310
00:15:12,730 --> 00:15:18,990
But I think they certainly
could help you to perfect

311
00:15:18,990 --> 00:15:22,850
or at least uncover some
of the obvious pitfalls

312
00:15:22,850 --> 00:15:24,890
in your arguments.

313
00:15:24,890 --> 00:15:29,360
They might help a
beginning lawyer or someone

314
00:15:29,360 --> 00:15:33,020
who's not an expert in an area
operate with higher expertise.

315
00:15:33,020 --> 00:15:34,490
I don't think
they're necessarily

316
00:15:34,490 --> 00:15:40,730
going to form the world's expert
on the topic of their expertise

317
00:15:40,730 --> 00:15:43,063
but I think they will enable.

318
00:15:43,063 --> 00:15:44,230
I think they can be helpful.

319
00:15:44,230 --> 00:15:45,470
That's a very
exciting potential.

320
00:15:45,470 --> 00:15:47,090
But they're-- but
they're not going--

321
00:15:47,090 --> 00:15:50,030
I don't-- my goal has not
been to replace the lawyer

322
00:15:50,030 --> 00:15:52,010
and it's certainly not
to replace the judge.

323
00:15:52,010 --> 00:15:55,660
I mean, I don't even consider
that an interesting question.

324
00:15:55,660 --> 00:15:56,430
But--

325
00:15:56,430 --> 00:15:56,930
Cool.

326
00:15:56,930 --> 00:15:59,000
So are there-- just
to-- are there any

327
00:15:59,000 --> 00:16:01,480
closing stories or
anecdotes that you'd

328
00:16:01,480 --> 00:16:05,450
like the tell about your
experiences [INAUDIBLE]??

329
00:16:05,450 --> 00:16:09,030
Well, maybe this comment that--

330
00:16:09,030 --> 00:16:12,290
so going back to my
interesting cases

331
00:16:12,290 --> 00:16:15,660
and going back to the
interconnections of knowledge,

332
00:16:15,660 --> 00:16:19,700
in fact that was influenced
a lot by my undergraduate

333
00:16:19,700 --> 00:16:22,130
experience being in the computer
science department where

334
00:16:22,130 --> 00:16:25,860
hypertext was being developed.

335
00:16:25,860 --> 00:16:28,820
And so I mean,
that's an influence

336
00:16:28,820 --> 00:16:30,590
on my way of
thinking that I guess

337
00:16:30,590 --> 00:16:33,470
is pretty pervasive but
it comes from something

338
00:16:33,470 --> 00:16:36,560
outside of mathematics
which was my supposedly home

339
00:16:36,560 --> 00:16:40,160
discipline, outside of AI which
is my practicing discipline.

340
00:16:40,160 --> 00:16:43,760
So I think maybe one of the
lessons learned is you know,

341
00:16:43,760 --> 00:16:48,410
it's a multifaceted
enterprise and you

342
00:16:48,410 --> 00:16:51,290
get a lot of information, a
lot of ideas from elsewhere.

343
00:16:51,290 --> 00:16:52,250
And it's a lot of fun.

344
00:16:52,250 --> 00:16:55,370
It's a lot of fun to learn new
things such as a new discipline

345
00:16:55,370 --> 00:16:56,750
like the law.

346
00:16:56,750 --> 00:16:58,610
There's lots of
stories but you know.

347
00:16:58,610 --> 00:17:00,560
Yeah.

348
00:17:00,560 --> 00:17:01,210
So.

349
00:17:01,210 --> 00:17:01,710
Great.

350
00:17:01,710 --> 00:17:02,750
Well thank you so much.

351
00:17:02,750 --> 00:17:04,310
It was my pleasure.

352
00:17:04,310 --> 00:17:05,500
OK.

353
00:17:05,500 --> 00:17:06,000